The magnitude of the error does not have anything to do how few bits are flipped. At $1M a mask set I would venture to say the chance is not taken. A piece of rubylith fell off of one of my mask sets in 1980 the only occurrence I have ever experienced of this problem.
Unfortunately at this moment i don´t have a/the references at hand (thought i had, but have to dig deeper in the archive to retrieve) and as i don´t know exactly about the content of JAES and IEEE seminal papers (beside knowing the publications by Lipshitz,Vanderkooy, Wannamaker, Wright, Gray and Stockham) i can´t comment if they already covered/mentioned that point.
Although i remember that Lipshitz/Vanderkooy/Wannamaker at least already discussed that only subtractive jitter is able to render the quantization error completely independent from the input signal. Nonsubtractive dither (TPDF) achieves to (simplification) let the error appear to be psychoacoustically like white noise.
The Authors relied in that regard on listening experiments and we know that there is usually no guarantuee for correctness, although the results seem to confirm that the noise is uncolored/unmodulated.
But there is a caveat given the assumptions (non overload region, quite high sampling rate in comparison to the input signal) of the reasoning.
There seems to be a dependence of the factor between input frequency and sampling frequeny (integer or not) and some authors found that stochastic dither achieved better results.
Use of noiseshaping further complicates the optimization process.
That´s what the meta-analysis roughly found overall; it might be that for todays hirez (means more bits and more bandwidth) the estimation of the mean parameter would be even higher. And it is just an estimation for the mean groupwise, so a lot of individuals will do much better or much more worse, that said under the assumption that no hard physiological lower barrier exists.
For good reasons Reiss emphasized that much more experimental work is needed.
Not sure if that is more important to Joe Consumer than the recommendation of someone in whos hearing ability he trusts.......
Yes, a file system is a database of sorts. My understanding is that MS has had plans off and on to get rid of their existing file systems and replace them with some kind of more general relational database, but they could never get it to work acceptably.
Regarding where directory information is stored, it depends on the particular file system. In some cases it has been located near the inner part of the disk to speed lookup performance. Obviously, with ram cached raids, that doesn't really matter anymore. However, the file system directory is still something distinct from the file data itself. For one consideration, my understanding would be that for Windows, file types are part of filename extensions, so they are not stored in the file itself, whereas Macs have used file forks inside the data file itself to contain that information. But, yes at some level an unformatted disk doesn't know how data will be stored on it. When it comes to GFS or other more modern file systems, I don't know offhand if directory data and file data are stored in the same blocks. If they aren't, I would argue that they are still distinct for most purposes.
It does, the masks are subject to an automatic control process, that has as an input precisely the design database. The control process is set with a threshold of detecting errors.
If you as a designer didn't hear about such errors, it doesn't mean they don't exist, it means they were not reported, or reported but considered non critical.
Flying pieces of rubylith were a different story, I don't know, I was just born then. There is no human intervention in the mask making process today, it is fully automated, from data input to the final control. At our foundry there is 1 (one) engineer in charge for the whole mask making process, from data handling to final check, including defect control and DRC. That's for a production of $16 billion worth of chips every year. He's entering the clean area perhaps once a year, if that.
My point being that whatever is written to or read from a disk or other stream is all subject to the same process including error correction in the widest sense. In this context it doesn't make a difference whether the bits are part of the file data or the file name or timestamp. On the lowest level, everything is written to and read from disk allocation blocks and that is where the data integrity stuff operates.
Jan
Okay, sure. At a low level that's true. But, somewhere above the lower levels, things are different and potential problems can reside there.
yes someone linked, quoted a bit earlier in this thread...
at some point the lack of good tests for any specific audible effect requires some degree of reliance on 'over bounding', magnifying/exaggerating errors then extrapolating
like listening for TPDF signal correlation at 8 bits - when the 'issue' is the claims of dither 'fail' at 16 bits - below the noise floor of nearly all recordings, home playback environments
at some point the lack of good tests for any specific audible effect requires some degree of reliance on 'over bounding', magnifying/exaggerating errors then extrapolating
like listening for TPDF signal correlation at 8 bits - when the 'issue' is the claims of dither 'fail' at 16 bits - below the noise floor of nearly all recordings, home playback environments
Last edited:
I meant to say, take a valid database and randomly flip one bit. There is no way a priori to tell how much damage (if any) this does. I've tried it for fun, an .exe usually freezes or segmentation faults. Some .jpg's break some have bad pixels, same with .pdf's some flag corrupt file some don't. A .wav is usually fine (a pop of unknown amplitude) unless you hit the header.
Last edited:
Here is a freeware audio file checker that works on windows:
http://www.vuplayer.com/other.php
Edit: I haven't used it personally, just something I came across.
http://www.vuplayer.com/other.php
Edit: I haven't used it personally, just something I came across.
Last edited:
I've experienced this with a backup of photographs, there were some strange anomalies in appearance in damaged files, but they threw no errors. I also found the same damages on the backup. Because of such infrequent access, the issues went unnoticed for what was presumably a long time.
My assumption is that this could also hold true for audio files.
I do also firmly believe in snapshots over duplicates to be able to roll back instead of pull from a backup which may already contain duplicates of those errors. This holds true for user errors as well as corruption issues.
That is true, since the polygons in the mask data are always described in a vector format (some e-beam pattern generators may convert that to a raster/pixel format, though), so it's not about turning on/off pixels, but about altering coordinates, which is almost guaranteed to have a catastrophic effect.
The story about jpeg is the same as for any compressed data; change a bit and the decompressor can either generate a wrong output (which may or may not appear as visible altered, or get out of sync and generate junk at the output. The highest the compression, the higher the risk of altering a bit leading to junk. Because of the fundamental source coding theorem, entropy is a limit to the length that we can use to lossless code a file and, at this limit, any bit change is catastrophic for the decompressor output.
If you look at the vector format as a lossless compressed format of a pixel/bitmap image, then it's now clear why the vector format is so much sensitive to changes.
Last edited:
I have had a music file get bit corruption, but only one. I was testing a proof of concept for a music server using some ropey old hardware I had lying around. Only this one track had a problem. Still played but with a small distorted segment. copied the damaged track off, deleted it and put a good copy on again, same issue. Then the hardware decided it had had enough of life and died. Lesson learned and I bought an HP microserver, which is the best £120 I have spent on muscial enjoyment in a long time.
But I kept the corrupted file and one day will load it into audacity to find out what has gone wrong. But that was operating hardware and software well out its comfort zone so the result was hardly suprising.
But I kept the corrupted file and one day will load it into audacity to find out what has gone wrong. But that was operating hardware and software well out its comfort zone so the result was hardly suprising.
Harmon Kardon's R&D Budget is $400 million! WOW!
I just saw this tidbit in the July 2017 issue of Stereophile (below).
Beats By Dr. Dre must have an R&D Budget of BILLIONS! Same with Bose and Magico and Wilson Audio and Vandersteen and Constellation and all the other companies that are kicking Harmon Kardon's asszs in the marketplace. Sonya Vabeetch!
_
I just saw this tidbit in the July 2017 issue of Stereophile (below).
Beats By Dr. Dre must have an R&D Budget of BILLIONS! Same with Bose and Magico and Wilson Audio and Vandersteen and Constellation and all the other companies that are kicking Harmon Kardon's asszs in the marketplace. Sonya Vabeetch!
_
Attachments
Harman International not Harmon Kardon. But it doesn't say what share of the budget. Total sales last reported of 7 billion dollars. So a bit more than the usual ratio of 5%. Of course Samsung expects the automotive market share to hit 100 billion over a bit under 10 years. So the lion's share probably goes to that market.
$400M. Dimitri should be in a good mood lately.
And as I read in the attachment, his hard work has been acknowledged.
George
It looks like we are in an impasse again. '-) I can hardly follow what you guys are debating about, but then I'm an old guy who doesn't really like digital playback, even now.
However, I got my digital playback going over the weekend to preview a sampler Blu-ray recorded at 24-96K. The music samples are varied and interesting, BUT the sound quality is marginal, even annoying. Now, is this because my equipment is not good enough (for me)? No, because I started my listening session with some Chesky 24-96K recordings that sounded more than acceptable with my digital playback. What is it? I think it is in the console electronics, but I can't be sure. That is the only consistent factor in the equation for EVERY selection, that I can't get around. Should I tell the record company? People usually do not appreciate 'feedback' like this, so I think I will just let it go.
However, I got my digital playback going over the weekend to preview a sampler Blu-ray recorded at 24-96K. The music samples are varied and interesting, BUT the sound quality is marginal, even annoying. Now, is this because my equipment is not good enough (for me)? No, because I started my listening session with some Chesky 24-96K recordings that sounded more than acceptable with my digital playback. What is it? I think it is in the console electronics, but I can't be sure. That is the only consistent factor in the equation for EVERY selection, that I can't get around. Should I tell the record company? People usually do not appreciate 'feedback' like this, so I think I will just let it go.
- Status
- Not open for further replies.